Abstract

Phenological monitoring of Actinidia chinensis is critical for optimising operational costs and yield prediction. However, current manual assessment methods are time-consuming, making them impractical for large-scale precision agriculture applications. Most existing phenological datasets focus exclusively on image data without spatial validation. The Multi-Modal Actinidia chinensis Phenology Dataset is composed of (i) 1 665 annotated images of phenological stages from bud to fruit set and (ii) georeferenced videos with systematic manual ground truth of spatial stage distributions. The dataset employs an adapted 17-class BBCH system that consolidates visually similar stages, excludes problematic categories, and introduces generic structural classes to address practical annotation difficulties. Additionally, the data is organised hierarchically across various plant structures, genders, and phenological stages. The annotated images offer versatility for a range of applications, including training data for computer vision models to detect phenological stages. Furthermore, the georeferenced videos facilitate the validation of automated counting algorithms. This combined approach enables plant-level detection accuracy and provides an illustrative methodology for spatial validation that users can extend to additional orchards, promoting the development and benchmarking of automated phenological monitoring systems for precision agriculture applications in kiwifruit production.

Abstract
The integration of Deep Learning techniques for grapevine yield estimation has led to significant advancements in Precision Viticulture. The accurate detection and counting of berries per bunch is a critical task that can explain up to 30% of yield variability, thereby enabling improved yield estimation. This study proposes a YOLO-based approach for the automated detection and counting of visible grapevine berries, using a dataset of more than 1500 images collected over three phenological stages. The selected YOLO models performed well in both detection and counting tasks, with all models achieving high detection accuracy (G-mAP ' 0.95) and estimation of visible berries (R2 ' 0.97). Among the evaluated models, YOLOv11n exhibited the highest detection performance (F1-Score = 0.954, G-mAP = 0.962), while YOLOv10n demonstrated the most consistent and reliable counting accuracy (MAPE = 4.764, MSE = 12.203, RMSE = 3.493). Beyond overall performance, the analysis revealed that ampelographic features such as berry size, occlusion, and bunch morphology can influence accuracy, although YOLOv10n showed no significant disparities across categories. To extend the scope, a complementary analysis demonstrated a strong linear relationship (R2 = 0.860) between visible counts and the total number of berries per bunch, supporting the potential of correction models to address occlusion. By systematically evaluating model behaviour across diverse viticultural conditions and incorporating correlation with total berry counts, this study provides a deeper understanding of the robustness and limitations of Deep Learning models, offering critical insights for future applications in vineyard monitoring, yield estimation, harvest optimisation, and management. © 2026 The Authors.

Abstract
In the evolving field of machine learning, ensuring group fairness has become a critical concern, prompting the development of algorithms designed to mitigate bias in decision-making processes. Group fairness refers to the principle that a model's decisions should be equitable across different groups defined by sensitive attributes such as gender or race, ensuring that individuals from privileged groups and unprivileged groups are treated fairly and receive similar outcomes. However, achieving fairness in the presence of group-specific concept drift remains an unexplored frontier, and our research represents pioneering efforts in this regard. Group-specific concept drift refers to situations where one group experiences concept drift over time, while another does not, leading to a decrease in fairness even if accuracy (ACC) remains fairly stable. Within the framework of federated learning (FL), where clients collaboratively train models, its distributed nature further amplifies these challenges since each client can experience group-specific concept drift independently while still sharing the same underlying concept, creating a complex and dynamic environment for maintaining fairness. The most significant contribution of our research is the formalization and introduction of the problem of group-specific concept drift and its distributed counterpart, shedding light on its critical importance in the field of fairness. In addition, leveraging insights from prior research, we adapt an existing distributed concept drift adaptation algorithm to tackle group-specific distributed concept drift, which uses a multimodel approach, a local group-specific drift detection mechanism, and continuous clustering of models over time. The findings from our experiments highlight the importance of addressing group-specific concept drift and its distributed counterpart to advance fairness in machine learning.

Abstract
This work addresses the joint challenges of layout design and power output smoothing in Airborne Wind Energy farms. The problem is deciding the power units' location and their synchronization, as well as balancing the power output in Airborne Wind Energy Farms, while maximizing the average power output. Airborne Wind Energy systems, particularly the ground-generation type, are susceptible to significant power output fluctuations, which must be mitigated to ensure a stable power flow injected into the electrical grid. We propose and evaluate various synchronization strategies, along with an innovative retraction method introduced in this work, which eliminates the need for synchronized operational cycles among kites. This approach increases flexibility in production cycle selection, leading to a significantly smoother power output signal.

Abstract

Abstract
Large Events Models (LEMs) are a class of models designed to predict and analyze the sequence of events in soccer matches, capturing the complex dynamics of the game. The original LEM framework, based on a chain of classifiers, faced challenges such as synchronization, scalability issues, and limited context utilization. This paper proposes a unified and scalable approach to model soccer events using a tabular autoregressive model. Our models demonstrate significant improvements over the original LEM, achieving higher accuracy in event prediction and better simulation quality, while also offering greater flexibility and scalability. The unified LEM framework enables a wide range of applications in soccer analytics that we display in this paper, including real-time match outcome prediction, player performance analysis, and game simulation, serving as a general solution for many problems in the field.